Graphical models are central to enterprise architecture and information systems work, yet they frequently fail to deliver value — not because the underlying analysis is wrong, but because the notations are poorly designed, mismatched to their audience, or unable to highlight what matters. This doctoral research paper sets out a programme of design science research aimed at improving visual language design and tooling, drawing on insights from human cognition, perception, semiotics, and graphic design. It introduces polymetric diagramming as a technique for making models more expressive and proposes a meta-meta model and tool architecture to support more effective visual language design and use.