Join MCMC kernels (simplify code), avoid Static Declaration of arrays (moved to shared or Global Memory) and reduce precision in some structures (double to float)