Implantable Cardioverter Defibrillators are the most common medical approach to prevent sudden cardiac death, which requires intracardiac electrograms to identify life-threatening ventricular arrhythmias. To accomplish reliable detection, current ventricular arrhythmias detection algorithms, however, depend on a number of heuristic detection criteria and frequently manual interventions to customize criteria parameters for each patient. In this paper, we present and explore the development of an ML approach for the detection of life-threatening Heart Arrhythmias through Intra-Cardiac Electrogram (IEGM) Data from an ICD Device. This work was facilitated by the design and analysis of 2 CNN models that could perform inference on a Low Power STM Nucleo-32 MCU. The first and second model design consisted of the construction of a 1D and 2D Convolutional Neural Network respectively. Multiple Microcontroller software platforms were utilized to construct and deploy the trained models onto the MCU platform for inference measurements. The experimental analysis consisted of minimizing Average Inference time and on board Memory Occupation while maximizing the accuracy of the models. With Convolutional Layer analysis and an emphasis on altering key CNN hyper parameters, our research examines and develops an informed report on the effect of Convolutional layers on the embedded device. Within the scope of our work is a thorough exploration of the differences between 1 Dimensional and 2 Dimensional CNN and their effect on the application of Neural Networks to MCU platforms.