It it possible to make a composite kernel?

[7sharp9]

Hi, I just came across this library but I wanted to know if it was easy to compose a composite kernel also where the RBF is anisotropic scaling.

I want to do something similar to what I have tested in scikit: Constant * GPR(RBF) with standard scaling on the input:

kernel = (ConstantKernel(constant_value=1.0, constant_value_bounds=(1e-2, 1e2)) *
          RBF(length_scale=[2.0, 6.0, 100.0], length_scale_bounds=(1e-2, 1e3)))

gpr = GaussianProcessRegressor(kernel=kernel, alpha=0.00036)

pipeline = Pipeline([
    ('scaler', StandardScaler()),
    ('gpr', gpr)
])

I have been using onnx but it's very slow, well in relative terms it's fast but ~100 microseconds it's still on the slow side when the input data is a relatively small dataset or ~100 rows of 3 columns.

Thanks!
Dave.