I forgot to mention that there is no need to match the spectral class of the reference star with the target star. The reason for choosing A stars, or if not a B star as a reference, is the spectra are very simple. So it is easy to smooth the response curve after dividing the spectra to get a curve that will act as a good response.
A good way to test a response curve is to take spectra of 2 A or B stars in the Miles database at similar airmass (altitude), using one to create the response curve, and then correct the other and compare it with the Miles database. A good fit means you are doing it right, though I still usually find they are not quite the same. Low resolution can be tricky.
I also think using the same reference stars is an excellent idea. It does need to be very close to the target so that its altitude stays similar as the 2 stars track across the sky.