I would suggest starting with hiring two bioinformaticians at the senior scientist level (e.g. ~10 yrs research experience) - one with expertise in proteomics, one with expertise in genomics. Look for former biologists who are talented on the programming/unix side, who can run their own desktops, a BAS and a data storage unit, but avoid hiring computer scientists or hardware specialists initially. Instead, rely on central IT services or cloud services for advanced HPC, and gear your bioinformatics unit to provide data analysis and research programming support.
I would skip hiring a director/manager in the first instance, but instead leave management to the academic group leader with substantial bioinformatics experience who has an established group in your institute. You probably won't have unlimited funds, so unless you hire a manager who can do one of the essential roles as well, you are just squandering cash that could be spent on someone actually doing the work. A small team doesn't need a manager, if you hire the right people.
Once you establish good bioinformatics services in your institute, this proto-team will be stretched to the limit. The next step would be to grow the group to have more junior members doing "blue color bioinformatics" - mapping reads, doing Mascot searches, etc. Hire these people from a CS background, with strong coding and documentation skills, but without a requirement to undestand the biology. Let the senior people train them in the biology on the job, and let them take over the automation from the senior core members (and clean up their rapidly developed script archive.)
Lastly, hire a senior person to do systems biology and data integration, who can also take over management of the group and devolve responsibility from the academic lead. The manager should be at (almost) a PI level in your institute, and should be the one who coordinates with users and other non-core bioinformaticians embedded in wet-lab groups. They should also be involved in faculty meetings so they have access to scientific and infrastructure plans, and can evolve the group to meet the needs of your institute over time. If you don't have this high-level academic-core interface, you risk treating your core as a technical pool, rather than the engine that will make your 'omics data collection actually pay off.
EDIT 5 July 2011:
Just recalled that there are a couple interesting perspectives on this issue published in PLoS Com Biol in addition to the Bioinformatics editorial noted by Stefano:
You also might want to check out some of the discussions of the Bioinfo-core group or other resources on their wiki.