Aspects are more cross-cutting concerns like security, memory management, multi-core or cloud, fault-tolerance, traceability etc. more likely to come into play when targeting some specific “platform” at model transformation time, hence a concern to be supported by some target-specific model compilers. (= not part of metalevel M0 domain analysis)
A DSL (or here a DSML) is defined and designed as a widely deployable and specialized modeling language (at M2 metalevel), focusing on supporting modeling of a certain domain (at M1 metalevel), like telecom network architectures, for instance.
You can check out our presentation at EclipseCon Europe to get the idea and big picture of it.
Your proprietary DSL (for instance) is deployed as DSL-plugins in Eclipse for wide use, supporting the UI customizations your DSL need, and your DS language model validation defined in OCL in your DSL.
I disagree with the statement that DSLs are small languages, focusing only on a particular aspect of a system, and that it only has to do with a certain domain of a system.
We must not forget needed UI customizations for your DSML. :)