using a pair of double* and an int for size is very antiquated and very c-style. there is probably no reason why cmaes intefarces should not use vector. this includes for instance the following:

CMAParameters<>(unsigned int n, const double* const p, const double sigma);

and the minimizer lambda

FitFunc Likelihood = [](const double* const p, const unsigned int n) { ... };

which could be e.g. simply

CMAParameters<>(const vector& p, const double sigma);
FitFunc Likelihood = [](const vector& p) { ... };