The early stages of development of real-time systems pose a big challenge due to having to meet strict requirements. This leads to engineers making assumptions and decisions based on their experience, which can be costly since certain issues will only show themselves in the testing phase when the project is almost entirely done. This leads to potentially costly redesigns in terms of both time and money. This thesis therefore proposes training predictive models to estimate CPU utilization based on the block-based programming language HiDraw, in use by Hitachi Energy for their MACH Control and Protection system. These models use symbols (blocks) to predict CPU utilization using various linear regression models (OLS, ridge, lasso, and elastic net) to plot a linear relationship between a symbol's occurrence and its contribution to the total CPU utilization of a program created in HiDraw. Due to unforeseen circumstances, there was not enough time to gather adequate data. Preliminary conclusions were drawn from the available data, however, the answer to the primary research questions remain inconclusive but suggestions for improvement were made for future work.