This paper explores the advantages of adding an MMX extension to the MicroBlaze (MB) soft processor,through its FSL ports. The implemented MMX unit is described in detail, were emphasis has been done in parallelizing the instruction execution. The propagation delays and FPGA resources consumed obtained for the different blocks during the synthesis phase for a Virtex-II target device are also shown. The structure of the peripheral itself, built by merging the MMX extension with a finite state machine and some additional glue logic is also described. In particular,attention is centered in the idea of using of a local micro-program memory which avoids the continuous transfer of codes to the peripheral. Finally, the method used to evaluate and compare the MB performance in calculating three well different algorithms with and without using the MMX extension, is also presented and discussed.