GP kernel dependent on latent variables

@bwengals re case 1 (tomography): I had a look at those covariance functions, but they depend only on X, I’d like to find a way to include some y(X) dependence as well. I’m just not sure the GP framework supports such a hierarchy.