Convolutional neural networks remain computationally demanding for deployment on edge devices. While pruning has emerged as an effective strategy for model compression, most existing methods rely on access to training data for retraining, which is often infeasible due to privacy, licensing, or storage constraints. We present Portable Pruning, a unified framework for data-free, structured, and unstructured pruning of convolutional neural networks without any access to training data. The framework incorporates \(L_1\) -based sparsity pruning, L \(_n\) -norm-based channel pruning, and their random baselines, enabling reproducible comparison across architectures and pruning granularities. A key contribution is our ablation analysis over compression ratios, which reveals how performance degrades under increasing sparsity, offering practical insights for deployment trade-offs. To support the community, we open-source the entire toolkit with automated benchmarking, ONNX export, and result logging. Our experiments on ResNet-18 and MobileNetV2 demonstrate that even simple data-free strategies can achieve competitive compression, supporting practical deployment under constrained conditions.