If you have a Windows application (rich client), then yes, certain bits and pieces might make sense to be loaded at startup (or when they're being used first), and then retained in memory. Things that you need frequently, and that don't change too much - things like information on your current user, or other things like that.
Linq-to-SQL is based on an existing database - it basically scan the database and creates classes from that. It doesn't really have a "model-first" domain-driven design approach. If you're more into creating your domain model first and deriving your database from that, I would recommend checking out the Entity Framework v4 that will ship with .NET 4.0 in March 2010 - that's quite specifically designed to do "model-first" / DDD scenarios.