UNCERTAINTY BOUNDS FOR KERNEL-BASED REGRESSION: A BAYESIAN SPS APPROACH