A typical way to proceed would be to use open source libraries (or create them, if needed). As you're probably aware, Intel supports performance libraries including IPP and MKL which have their own forums neighboring this one, and are included with compiler distributions.
Prior to writing assembly code, it's probably useful to make representative benchmarks and profile them with open source tools such as gprof, maybe oprofile, and Intel tools such as VTune Amplifier. If you find specific sections of code where you want expert suggestions, you can get help on appropriate language forums, including the neighbor forums of this one.
If your applications gain sufficient acceptance to become a significant part of equipment selection criteria, or if the value to your company justifies engaging a consultant, further steps could be considered.
