This pull request implements a simple version of Gaussian Process Regression and RBF kernel under the pyro.contrib.gp module. Comparing to other GP packages, it lacks a huge number of features. So things to do in the future might be

• Add more kernels, support adding or multiplying kernels together.
• Add likelihood, mean_function modules.
• Add other models (such as classification, Poisson regression, dimensional reduction).
• Support spare methods (using inducing points).
• Support multidimensional output learning.
• ... (due to the lack of my knowledge, I can't name them here).