How to go from Simulink to embedded fixed-point DSP processor?

A library of audio algorithms is modeled, tested and verified in Simulink. It needs to go from this existing Simulink models down to a multiple embedded platforms (both floating and fixed-point DSP processors: Qualcomm, Tensilica, ARM, etc).

Several workflows are considered:

1) Simulink diagram -> Matlab code -> generic C floating point -> C embedded platform 1
                                                               -> C embedded platform 2
                                                               -> C embedded platform 3

2) Simulink diagram -> generic C floating point -> C embedded platform 1
                                                -> C embedded platform 2
                                                -> C embedded platform 3

3) Simulink diagram -> C embedded platform 1
                       C embedded platform 2
                       C embedded platform 3

The Simulink models keep changing, so versioning and ease of maintenance should be considered.

Which workflow should be preferred?

Note: Simulink has a C code generation feature. It can generate both generic and embedded C code (with some of the required embedded platforms supported (ARM), and some not (Tensilica)). Not sure about the quality and usability of the C code, so should code generation be used or not is also an important question.