I'd just use a PCR fragment, in all honesty. Even starting with genomic DNA, the overwhelming majority of your amplification events are going to be from primers binding to PCR fragments, so you might as well just start with that. It's also much easier to obtain large amounts of PCR fragment.
I typically run my PCR, run that on a gel, gel extract the band so I know that my measured [DNA] pertains only to fragment (and not primers etc), measure the concentration and then (via comparison with length of fragment) obtain molecules.ul-1, and then just use this for a standard curve, from around 10^8 molecules.well-1 down to 10^-1 or 10^-2.
At the lower values (10, 1, 0.1 or 0.01 templates per well) you should see increasingly stochastic behaviour (yes/no signals, or wildly varied Cq values): this is fine, as this is what stochastic partitioning would suggest. The entire curve should thus allow you to not only calculate copy numbers, but also give you a clear idea of which values represent the threshold for trustworthy data.