On the use of kernel-based methods in sound synthesis by physical modeling