Have you check out the discussion in this post? Multiple (uncertain) function observations of the same Gaussian process