A Kernel-based Discrimination Framework for Solving Hypothesis Testing Problems with Application to Speaker Verification