Parallel Gaussian process with kernel approximation in CUDA