Log-concavity Results on Gaussian Process Methods for Supervised and Unsupervised Learning