You might design two primer sets with the 5' end of one amplicon (PCR product) at the deletion position, and the 3' end of the other amplicon adjacent to the deletion. Then extend the 5' end of one primer bordering on the deletion (the reverse primer of the upstream set, or the forward primer of the downstream primer set) to include sequence complementary to the the other primer, in other words extend the 5' end of the primer skipping the entire deletion and continuing to contain complementary sequence to the primer on the far side of the deletion. (Draw it out. Do you understand?) This only needs to be done to one primer or the other flanking the deletion and it will create PCR products with the same overlap as one of the normal primers, so the overlap extension can proceed like a normal primer anneal and extension. So, the two PCR products are made, purified, then a small amount of each is mixed together in a final PCR reaction with only the outer primers.
I think this is what you were asking. An alternative would be to do a mutagenesis type reaction with a pair of complementary primers and recreate an entire plasmid without having to ligate an insert into it. This is less sure to work than a point mutagenesis, but I think I had it work a couple of times, and fail once.