SIMD does not really make sense for the found candidates and would only uglyfy the code.
Offering special iterators for Kernel is also problematic since the current for_each implementation still enables the programmer to access different values.
For Kernel there is already SIMD support, so the case we were initially targetting was solved. Tacking on SIMD for the sake of it would not produce worthwhile speedups for the more complex code and additionl indirections (which might be zero cost but we also optimize for ease of use).
The steps that will be taken are