[GPU] Fix mvn segmentation fault for multiple threads start_async (#2…

…7654)

### Details:
- *Deep copy impl params*: deep copy primitive impl KernelData params
for different networks when running start_async() with multi-threads
(THROUGHPUT mode)

### Tickets:
 - *[CVS-155564](https://jira.devtools.intel.com/browse/CVS-155564)*

Migrated from PR #27263.

---------

Signed-off-by: yuan.xiong <[email protected]>

src/plugins/intel_gpu/src/graph/impls/ocl/activation.cpp

@@ -28,7 +28,7 @@ struct activation_impl : typed_primitive_impl_ocl<activation> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::activation_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<activation_impl>(*this);
+        return make_deep_copy<activation_impl, kernel_params_t>(*this);
     }
 
     void load(BinaryInputBuffer& ib) override {

src/plugins/intel_gpu/src/graph/impls/ocl/adaptive_pooling.cpp

@@ -19,7 +19,7 @@ struct adaptive_pooling_impl : public typed_primitive_impl_ocl<adaptive_pooling>
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::adaptive_pooling_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<adaptive_pooling_impl>(*this);
+        return make_deep_copy<adaptive_pooling_impl, kernel_params_t>(*this);
     }
 
 protected:

src/plugins/intel_gpu/src/graph/impls/ocl/arg_max_min.cpp

@@ -42,7 +42,7 @@ struct arg_max_min_impl : typed_primitive_impl_ocl<arg_max_min> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::arg_max_min_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<arg_max_min_impl>(*this);
+        return make_deep_copy<arg_max_min_impl, kernel_params_t>(*this);
     }
 
     void load(BinaryInputBuffer& ib) override {

src/plugins/intel_gpu/src/graph/impls/ocl/batch_to_space.cpp

@@ -18,7 +18,7 @@ struct batch_to_space_impl : typed_primitive_impl_ocl<batch_to_space> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::batch_to_space_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<batch_to_space_impl>(*this);
+        return make_deep_copy<batch_to_space_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param) {

src/plugins/intel_gpu/src/graph/impls/ocl/border.cpp

@@ -21,7 +21,7 @@ struct border_impl : typed_primitive_impl_ocl<border> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::border_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<border_impl>(*this);
+        return make_deep_copy<border_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param, bool is_shape_agnostic = false) {

src/plugins/intel_gpu/src/graph/impls/ocl/broadcast.cpp

@@ -20,7 +20,7 @@ struct broadcast_impl : typed_primitive_impl_ocl<broadcast> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::broadcast_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<broadcast_impl>(*this);
+        return make_deep_copy<broadcast_impl, kernel_params_t>(*this);
     }
 
     void load(BinaryInputBuffer& ib) override {

src/plugins/intel_gpu/src/graph/impls/ocl/bucketize.cpp

@@ -20,7 +20,7 @@ struct bucketize_impl : typed_primitive_impl_ocl<bucketize> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::bucketize_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<bucketize_impl>(*this);
+        return make_deep_copy<bucketize_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param) {

src/plugins/intel_gpu/src/graph/impls/ocl/concatenation.cpp

@@ -50,7 +50,7 @@ struct concatenation_impl : typed_primitive_impl_ocl<concatenation> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::concatenation_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<concatenation_impl>(*this);
+        return make_deep_copy<concatenation_impl, kernel_params_t>(*this);
     }
 
     void load(BinaryInputBuffer& ib) override {

src/plugins/intel_gpu/src/graph/impls/ocl/convert_color.cpp

@@ -19,7 +19,7 @@ struct convert_color_impl : typed_primitive_impl_ocl<convert_color> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::convert_color_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<convert_color_impl>(*this);
+        return make_deep_copy<convert_color_impl, kernel_params_t>(*this);
     }
 
 protected:

src/plugins/intel_gpu/src/graph/impls/ocl/convolution.cpp

@@ -24,7 +24,7 @@ struct convolution_impl : typed_primitive_impl_ocl<convolution> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::convolution_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<convolution_impl>(*this);
+        return make_deep_copy<convolution_impl, kernel_params_t>(*this);
     }
 
     void load(BinaryInputBuffer& ib) override {

src/plugins/intel_gpu/src/graph/impls/ocl/crop.cpp

@@ -20,7 +20,7 @@ struct crop_impl : typed_primitive_impl_ocl<crop> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::crop_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<crop_impl>(*this);
+        return make_deep_copy<crop_impl, kernel_params_t>(*this);
     }
 
     void load(BinaryInputBuffer& ib) override {

src/plugins/intel_gpu/src/graph/impls/ocl/ctc_greedy_decoder.cpp

@@ -32,7 +32,7 @@ struct ctc_greedy_decoder_impl : typed_primitive_impl_ocl<ctc_greedy_decoder> {
 
 public:
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<ctc_greedy_decoder_impl>(*this);
+        return make_deep_copy<ctc_greedy_decoder_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param) {

src/plugins/intel_gpu/src/graph/impls/ocl/ctc_loss.cpp

@@ -20,7 +20,7 @@ struct ctc_loss_impl : typed_primitive_impl_ocl<ctc_loss> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::ctc_loss_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<ctc_loss_impl>(*this);
+        return make_deep_copy<ctc_loss_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param) {

src/plugins/intel_gpu/src/graph/impls/ocl/cum_sum.cpp

@@ -53,7 +53,7 @@ struct cum_sum_impl : typed_primitive_impl_ocl<cum_sum> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::cum_sum_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<cum_sum_impl>(*this);
+        return make_deep_copy<cum_sum_impl, kernel_params_t>(*this);
     }
 
     void load(BinaryInputBuffer& ib) override {

src/plugins/intel_gpu/src/graph/impls/ocl/deconvolution.cpp

@@ -21,7 +21,7 @@ struct deconvolution_impl : typed_primitive_impl_ocl<deconvolution> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::deconvolution_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<deconvolution_impl>(*this);
+        return make_deep_copy<deconvolution_impl, kernel_params_t>(*this);
     }
 
 protected:

src/plugins/intel_gpu/src/graph/impls/ocl/depth_to_space.cpp

@@ -19,7 +19,7 @@ struct depth_to_space_impl : typed_primitive_impl_ocl<depth_to_space> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::depth_to_space_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<depth_to_space_impl>(*this);
+        return make_deep_copy<depth_to_space_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param) {

src/plugins/intel_gpu/src/graph/impls/ocl/detection_output.cpp

@@ -21,7 +21,7 @@ struct detection_output_impl : typed_primitive_impl_ocl<detection_output> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::detection_output_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<detection_output_impl>(*this);
+        return make_deep_copy<detection_output_impl, kernel_params_t>(*this);
     }
 
 public:

src/plugins/intel_gpu/src/graph/impls/ocl/dft.cpp

@@ -19,7 +19,7 @@ struct dft_impl : typed_primitive_impl_ocl<dft> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::dft_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<dft_impl>(*this);
+        return make_deep_copy<dft_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param) {

src/plugins/intel_gpu/src/graph/impls/ocl/dynamic_quantize.cpp

@@ -21,7 +21,7 @@ struct dynamic_quantize_impl : typed_primitive_impl_ocl<dynamic_quantize> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::dynamic_quantize_impl);
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<dynamic_quantize_impl>(*this);
+        return make_deep_copy<dynamic_quantize_impl, kernel_params_t>(*this);
     }
 
     void load(BinaryInputBuffer& ib) override {

src/plugins/intel_gpu/src/graph/impls/ocl/eltwise.cpp

@@ -20,7 +20,7 @@ struct eltwise_impl : typed_primitive_impl_ocl<eltwise> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::eltwise_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<eltwise_impl>(*this);
+        return make_deep_copy<eltwise_impl, kernel_params_t>(*this);
     }
 
     void load(BinaryInputBuffer& ib) override {

src/plugins/intel_gpu/src/graph/impls/ocl/embedding_bag.cpp

@@ -19,7 +19,7 @@ struct embedding_bag_impl : typed_primitive_impl_ocl<embedding_bag> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::embedding_bag_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<embedding_bag_impl>(*this);
+        return make_deep_copy<embedding_bag_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param) {

src/plugins/intel_gpu/src/graph/impls/ocl/experimental_detectron_detection_output.cpp

@@ -20,7 +20,7 @@ struct experimental_detectron_detection_output_impl
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::experimental_detectron_detection_output_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<experimental_detectron_detection_output_impl>(*this);
+        return make_deep_copy<experimental_detectron_detection_output_impl, kernel_params_t>(*this);
     }
 
 protected:

src/plugins/intel_gpu/src/graph/impls/ocl/experimental_detectron_generate_proposals_single_image.cpp

@@ -20,7 +20,7 @@ struct experimental_detectron_generate_proposals_single_image_impl
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::experimental_detectron_generate_proposals_single_image_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<experimental_detectron_generate_proposals_single_image_impl>(*this);
+        return make_deep_copy<experimental_detectron_generate_proposals_single_image_impl, kernel_params_t>(*this);
     }
 
 protected:

src/plugins/intel_gpu/src/graph/impls/ocl/experimental_detectron_prior_grid_generator.cpp

@@ -21,7 +21,7 @@ struct experimental_detectron_prior_grid_generator_impl
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::experimental_detectron_prior_grid_generator_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<experimental_detectron_prior_grid_generator_impl>(*this);
+        return make_deep_copy<experimental_detectron_prior_grid_generator_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param) {

src/plugins/intel_gpu/src/graph/impls/ocl/experimental_detectron_roi_feature_extractor.cpp

@@ -19,7 +19,7 @@ struct experimental_detectron_roi_feature_extractor_impl : public typed_primitiv
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::experimental_detectron_roi_feature_extractor_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<experimental_detectron_roi_feature_extractor_impl>(*this);
+        return make_deep_copy<experimental_detectron_roi_feature_extractor_impl, kernel_params_t>(*this);
     }
 
     event::ptr execute_impl(const std::vector<event::ptr>& events,

src/plugins/intel_gpu/src/graph/impls/ocl/experimental_detectron_topk_rois.cpp

@@ -20,7 +20,7 @@ struct experimental_detectron_topk_rois_impl : typed_primitive_impl_ocl<experime
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::experimental_detectron_topk_rois_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<experimental_detectron_topk_rois_impl>(*this);
+        return make_deep_copy<experimental_detectron_topk_rois_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param) {

src/plugins/intel_gpu/src/graph/impls/ocl/extract_image_patches.cpp

@@ -30,7 +30,7 @@ struct extract_image_patches_impl : typed_primitive_impl_ocl<extract_image_patch
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::extract_image_patches_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<extract_image_patches_impl>(*this);
+        return make_deep_copy<extract_image_patches_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param) {

src/plugins/intel_gpu/src/graph/impls/ocl/eye.cpp

@@ -20,7 +20,7 @@ struct eye_impl : typed_primitive_impl_ocl<eye> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::eye_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<eye_impl>(*this);
+        return make_deep_copy<eye_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param) {

src/plugins/intel_gpu/src/graph/impls/ocl/fully_connected.cpp

@@ -45,7 +45,7 @@ struct fully_connected_impl : typed_primitive_impl_ocl<fully_connected> {
     }
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<fully_connected_impl>(*this);
+        return make_deep_copy<fully_connected_impl, kernel_params_t>(*this);
     }
 
     void load(BinaryInputBuffer& ib) override {

src/plugins/intel_gpu/src/graph/impls/ocl/gather.cpp

@@ -64,7 +64,7 @@ struct gather_impl : typed_primitive_impl_ocl<gather> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::gather_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<gather_impl>(*this);
+        return make_deep_copy<gather_impl, kernel_params_t>(*this);
     }
 
     void load(BinaryInputBuffer& ib) override {

src/plugins/intel_gpu/src/graph/impls/ocl/gather_elements.cpp

@@ -51,7 +51,7 @@ struct gather_elements_impl : typed_primitive_impl_ocl<gather_elements> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::gather_elements_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<gather_elements_impl>(*this);
+        return make_deep_copy<gather_elements_impl, kernel_params_t>(*this);
     }
 
     void load(BinaryInputBuffer& ib) override {

src/plugins/intel_gpu/src/graph/impls/ocl/gather_nd.cpp

@@ -21,7 +21,7 @@ struct gather_nd_impl : typed_primitive_impl_ocl<gather_nd> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::gather_nd_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<gather_nd_impl>(*this);
+        return make_deep_copy<gather_nd_impl, kernel_params_t>(*this);
     }
 
     void load(BinaryInputBuffer& ib) override {

src/plugins/intel_gpu/src/graph/impls/ocl/gather_tree.cpp

@@ -20,7 +20,7 @@ struct gather_tree_impl : typed_primitive_impl_ocl<gather_tree> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::gather_tree_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<gather_tree_impl>(*this);
+        return make_deep_copy<gather_tree_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param) {

src/plugins/intel_gpu/src/graph/impls/ocl/gemm.cpp

@@ -32,7 +32,7 @@ struct gemm_impl : multi_stage_primitive<gemm> {
     const uint32_t indirect_gemm = 1;
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<gemm_impl>(*this);
+        return make_deep_copy<gemm_impl, kernel_params_t>(*this);
     }
 
     gemm_impl() = default;

src/plugins/intel_gpu/src/graph/impls/ocl/generate_proposals.cpp

@@ -20,7 +20,7 @@ struct generate_proposals_impl
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::generate_proposals_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<generate_proposals_impl>(*this);
+        return make_deep_copy<generate_proposals_impl, kernel_params_t>(*this);
     }
 
 protected:

src/plugins/intel_gpu/src/graph/impls/ocl/grid_sample.cpp

@@ -49,7 +49,7 @@ struct grid_sample_impl : public typed_primitive_impl_ocl<grid_sample> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::grid_sample_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<grid_sample_impl>(*this);
+        return make_deep_copy<grid_sample_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param) {

src/plugins/intel_gpu/src/graph/impls/ocl/grn.cpp

@@ -20,7 +20,7 @@ struct grn_impl : typed_primitive_impl_ocl<grn> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::grn_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<grn_impl>(*this);
+        return make_deep_copy<grn_impl, kernel_params_t>(*this);
     }
 
 public:

src/plugins/intel_gpu/src/graph/impls/ocl/group_normalization.cpp

@@ -18,7 +18,7 @@ struct group_normalization_impl : typed_primitive_impl_ocl<group_normalization>
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::group_normalization_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<group_normalization_impl>(*this);
+        return make_deep_copy<group_normalization_impl, kernel_params_t>(*this);
     }
 
     void load(BinaryInputBuffer& ib) override {

src/plugins/intel_gpu/src/graph/impls/ocl/kv_cache.cpp

@@ -68,7 +68,7 @@ struct kv_cache_impl : multi_stage_primitive<kv_cache> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::kv_cache_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<kv_cache_impl>(*this);
+        return make_deep_copy<kv_cache_impl, kernel_params_t>(*this);
     }
 
     const size_t concat_stage = 0;

src/plugins/intel_gpu/src/graph/impls/ocl/lrn.cpp

@@ -20,7 +20,7 @@ struct lrn_impl : typed_primitive_impl_ocl<lrn> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::lrn_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<lrn_impl>(*this);
+        return make_deep_copy<lrn_impl, kernel_params_t>(*this);
     }
 
     static kernel_params_t get_kernel_params(const kernel_impl_params& impl_param) {

src/plugins/intel_gpu/src/graph/impls/ocl/lstm_elt.cpp

@@ -20,7 +20,7 @@ struct lstm_elt_impl : typed_primitive_impl_ocl<lstm_elt> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::lstm_elt_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<lstm_elt_impl>(*this);
+        return make_deep_copy<lstm_elt_impl, kernel_params_t>(*this);
     }
 
 protected:

src/plugins/intel_gpu/src/graph/impls/ocl/matrix_nms.cpp

@@ -44,7 +44,7 @@ struct matrix_nms_impl : typed_primitive_impl_ocl<matrix_nms> {
     DECLARE_OBJECT_TYPE_SERIALIZATION(cldnn::ocl::matrix_nms_impl)
 
     std::unique_ptr<primitive_impl> clone() const override {
-        return make_unique<matrix_nms_impl>(*this);
+        return make_deep_copy<matrix_nms_impl, kernel_params_t>(*this);
     }
 
 protected:

src/plugins/intel_gpu/src/graph/impls/ocl/multi_stage_primitive.hpp

@@ -141,6 +141,18 @@ struct multi_stage_primitive : public typed_primitive_impl<PType> {
         return {kernels_cache.get_cached_kernel_ids(_kernels)};
     }
 
+    template<typename ImplType, typename KernelParamsType>
+    static std::unique_ptr<primitive_impl> make_deep_copy(const ImplType& impl_ocl) {
+        auto prim_impl = make_unique<ImplType>(impl_ocl);
+        for (auto& _kernel_data : (*prim_impl)._kernels_data) {
+            KernelParamsType* params_ptr = dynamic_cast<KernelParamsType*>(_kernel_data.params.get());
+            if (params_ptr != nullptr) {
+                _kernel_data.params = make_unique<KernelParamsType>(*params_ptr);
+            }
+        }
+        return prim_impl;
+    }
+
     std::vector<kernel::ptr> get_kernels() const override {
         return _kernels;
     }