Have a look at SAGE (https://en.wikipedia.org/wiki/Serial_analysis_of_gene_expres...), might give you some ideas for joining short reads. I suppose you'd want to look for a restriction enzyme that creates average fragment size in your desired range.