These days, there is a growing industrial need for more flexible methodologies in the field of model-driven software development. In the development phase, it has become more essential to follow the modification of the specification by making changes in the model dynamically. There is a wide range of mainstream metamodeling methodologies which support the two-level approach. These methodologies do not meet all of the requirements of the software industry because they are not flexible enough and it can be problematic to add additional elements or relationships to the particular model. Multi-level metamodeling can improve flexibility since the modelling levels can be compatible with the abstraction levels of the given domain problem. At the same time, multi-level metamodeling also raises a number of new issues, such as the need to rethink the instance-type relationship itself.
In the past decade, several approaches have been published in the literature that aimed to develop and refine the principles of multilevel metamodeling. At the same time, however, there are no widely accepted principles in the field, all approaches support different features.
The Dynamic Multi-Layer Algebra (DMLA) approach supports numerous new features which can grant a solution for many industrial problem settings. The DMLA approach is currently in a research phase, thus it is needed to test and analyze the features of DMLA.
In this thesis, I present the problems of multi-level metamodeling in general and I also present the modelling formalism of DMLA. I introduce a classification-based systematic approach which can facilitate the design of domain-specific visual languages. I use this approach to elaborate a visual language which can facilitate the process of multi-level metamodeling within DMLA. I also analyze the advantages and drawbacks of the most relevant approaches in the field of multi-level metamodeling by creating case studies. Finally, I summarize the results achieved and future plans related to DMLA.