Portable and embedded devices form an increasingly large group of computers, often referred to as Ambient Intelligence (AmI). This new variety in computing platforms will cause a corresponding diversity in software/hardware platforms and other context factors. Component-based middleware platforms offer a uniform environment for software, but they do not take away specific context differences, such as hardware resources, user identity/role and logical/physical location. Specialised component versions and/or configurations have to be made for each computing context if that computing context is to be used to its full extent. This is because the fine differences between component versions cannot be separated into finer components with the current component models. Aspect-oriented programming and generative programming technologies can be used to provide the fine-grained modularity that is necessary. In addition, the diversity of component-based platforms themselves form an extra reason for different component versions. We propose using a context-driven framework for the development of AmI components, which is based upon a gradual refinement mechanism. This refinement mechanism can cope with the course-grained differences between component models as well as the fine-grained differences between computing configurations.