Graphical representations such as circuit schematics, Karnaugh–Veitch (KV) maps, or Unied Modeling Language (UML) diagrams are widely used in computer science and engineering education. However, these visual formats are often inaccessible to blind and visually impaired (BVI) students, since they rely on visual perception for conveying both structure and semantics. Existing accessibility guidelines (e.g., WCAG) provide little support for subject-specific materials, leaving educators without actionable workflows for producing equivalent alternatives. We contribute a collection of five pedagogical design patterns that enable accessible representations of diagrams, graphics, and numerical data in Science, Technology, Engineering, and Mathematics (STEM) teaching. The patterns address (1) transformable text-first authoring, (2) accessible diagram generation via textual modeling, (3) tactile and haptic rendering workflows, (4) audio-based representations of time-series and quantitative plots, and (5) human-in-the-loop generation of explanatory descriptions. Each pattern is derived from classroom practice, supplemented by tool recommendations and implementation examples. Together, the collection provides a structured methodology for producing multi-modal STEM learning materials that are usable with screen readers, Braille displays, tactile printers, or sonification tools. While developed in the context of computer engineering education, the patterns generalize to other domains that rely on diagrammatic reasoning.