We consider a methodology for flexible software design, runtime programming, defined by recurrent, incremental software modifications to a program at runtime, called runtime patches. The principles we consider for runtime programming are model preservation and scalability. Model preservation means that a runtime patch preserves the programming model in place for programs -- in terms of syntax, semantics, and correctness properties -- as opposed to an ``ad-hoc'', disruptive operation, or one that requires an extra level of abstraction. Scalability means that, for practicality and performance, the effort in program compilation required by a runtime patch should ideally scale in proportion to the change induced by it. We formulate runtime programming over an abstract model for component-based concurrent programs, defined by a modular relation between the syntax and semantics of programs, plus built-in notions of initialization and quiescence. The notion of a runtime patch is defined over these assumptions, as a model-preserving transition between two programs and respective states. Additionally, we propose an incremental compilation framework for scalability in patch compilation. The formulation is put in perspective through a case-study instantiation over a language for distributed hard real-time systems, the Hierarchical Timing Language (HTL).