Usually, the development of DSMLs is separated from the development of corresponding models, resulting in two different representations that are not integrated. Hence, it is not possible to modify a language, while it is being used. Also, DSMLs are usually specified from scratch with some generic meta-modeling language, even though it would often be more appropriate to use a further, less specific DSML for that purpose – similar to the evolution of technical languages. The lack of expressivenes of languages used to specify DSMLs is a serious obstacle to representing all relevant knowledge about the targeted domain without redundancy. Also, common approaches to modeldriven development are based on separate representations of models and code, which makes keeping them synchronized extremely challenging. These obstacles are a consequence of principal limitations imposed by MOF and architectures of prevailing programming languages. In this tutorial it is demonstrated how these obstacles can be overcome by the use of a multi-level language architecture that allows for a common representation of languages, models and programs, which, in addition, enables software systems that are integrated with their underlying conceptual models at runtime. The respective language engineering and execution environment, XModelerML©, can be downloaded from www.le4mm.org.