Opening the Black Box: Revealing Interpretable Sequence Motifs in Kernel-Based Learning Algorithms