I suggest you search Google and/or PubMed for the terms "copy number variation" and "simulated data." There is a lot of literature on CNV detection and the authors often generate simulated data in order to test their algorithms. A good article will give details of the simulation process.
@DK: Sorry! I didn't give enough information, simulating paired-end reads.
I inserted and deleted some regions of fasta files to make CNVs of genes; after that Wgsim to simulate.
But I'd like to look for a software making real CNVs, especially: heterogeneous copy numbers.