Application of Kernel-Based Feature Space Transformations and Learning Methods to Phoneme Classification