[c++]float[16]

You could define your vector as
vector vVect;

and then insert either like:
float *matrix = new float[16];
// initialize matrix
vVect.push_back(matrix);

or
float matrix[16] = { values };
vVect.push_back((float *)&matrix);

Don't forget to delete the pointers when you are done with the vector though, possibly in the destructor of the class that has this vector as a class member.

I'd suggest you write a Matrix class to encapsulate the data and operations related to it though. You could provide methods in that class to get the pointer to the raw data as well, so it would make no difference and be more convenient to read and write code.

Arrays are not first-class types. You can't really store them like that directly. You can make pointers to new allocations and store the pointers in the vector, but if this is the only reason you're doing it, then that (a) wastes memory, (b) may hurt performance, and (c) defeats the purpose of using the standard library container in the first place. (One slightly better reason to use the pointers is if the "arrays" you wanted to store can be different sizes within the same vector.)

The simplest thing is to wrap the array in some kind of structure. This work has already been done for you.

I always found it kind of funny that you could not assign arrays directly, but when you put them in a struct, it suddenly works :)