Cancer is one of the top leading causes of death worldwide and the most common type of cancer amongst women is breast cancer. Breast cancer treatment can be highly effective if the cases are detected and diagnosed early, and the best way to detect the presence of breast cancer with certainty is by conducting a biopsy. The tissue specimens examined for diagnosis are captured with different magnifications, as the most informative magnification is still undecided amongst pathologists. Automating breast cancer detection through Computer-Aided Diagnosis (CAD) is a very crucial task, given the importance of early detection for proper treatment and the workload shortage of pathologists meant to carry out the task manually. Amongst the different CAD methods that have been proposed, Convolutional Neural Networks (CNN) achieved great success in histopathology image analysis as of late. Unfortunately, handcrafting a CNN architecture that suits a specific dataset is a specialized trial-and error task that is tedious and time-consuming. An alternative to manually designing a CNN is Neural Architecture Search (NAS) methods, which automatically design optimized CNN-based solutions. However, NAS methods tend to require a lot of time and/or powerful computing resources. The NAS-Bench-101 dataset was recently created to make NAS methods more accessible to researchers for benchmarking and experimenting with different NAS methods. In this work, we propose to classify breast cancer histopathology images independently of their magnification with a CNN designed automatically with the use of NAS-Bench 101 and the NAS search algorithm Regularized Evolution. We have proposed two architectures, a single-task network that predicts malignancy and a muli-task network that predicts both malignancy and magnification. The networks were evaluated on the BreakHis dataset, and the results are comparable to previous state-of-the-art methods that are magnification-specific and surpass previous magnification-independent approaches.