Building conceptual models for software design, in particular for high-tech applications such as smart home systems, is a complex task that significantly affects the efficiency of their development processes. One of the innovative methods of solving this problem is the use of domain-specific modeling languages (DSMLs), which can reduce the time and other project resources required to create such systems. The subject of research in this paper is approaches to the development of DSML for Smart Home systems as a separate class of Internet of Things systems. The purpose of this work is to propose an approach to the development of DSMLs based on a model of variability of the properties of such a system. The following tasks are being solved: analysis of some existing approaches to the creation of DSMLs; construction of a multifaceted classification of requirements for them, application of these requirements to the design of the syntax of a specific DSML-V for the creation of variable software in smart home systems; development of a technological scheme and quantitative metrics for experimental evaluation of the effectiveness of the proposed approach. The following methods are used: variability modeling based on the property model, formal notations for describing the syntax of the DSML-V language, and the use of the open CASE tool metaDepth. Results: a multifaceted classification of requirements for a broad class of DSML languages is built; the basic syntactic constructions of the DSML-V language are developed to support the properties of software variability of "Smart Home" systems; a formal description of such syntax in the Backus-Naur notation is given; a technological scheme for compiling DSML-V specifications into the syntax of the language of the open CASE tool metaDepth is created; the effectiveness of the proposed approach using quantitative metrics is experimentally investigated. Conclusions: the proposed method of developing a specialized problem-oriented language for smart home systems allows for multilevel modeling of the variability properties of its software components and provides an increase in the efficiency of programming such models by about 14% compared to existing approaches.