The growing volume and consistency of online ordering has renewed focus on the optimization of logistics and interest in technology-driven supply chain operations. Recent trials have demonstrated the viability of using unmanned aerial vehicles (UAVs), known colloquially as drones, in last-mile deliveries. In minimizing delivery times for customers, reducing the load of logistics workers, and reducing congestion and pollution per delivery, the integration of drones into traditional delivery networks presents transformative potential to benefit consumers, firms, and society at large. This dissertation investigates operational and tactical problems that inform routing and assignment decisions for vehicle-drone delivery systems. In three essays that follow, mathematical programming models are developed for these problems and their computational tractability enhanced via optimization constructs spanning preprocessing, valid inequalities, and reformulation techniques. Moreover, insights are provided into the network topographies that stand to benefit from the integration of drones into last-mile deliveries. The impact of these optimization techniques is assessed using benchmark instances from the growing literature. Chapter 1, based on the work of El-Adle et al. (2019), investigates the operational problem of assigning customers to be visited either by a delivery vehicle or by a portable companion drone launched from the vehicle, recently known in the literature as a Traveling Salesman Problem with Drone (TSP-D). A novel 0-1 mixed-integer program (MIP) is proposed for the TSP-D that synchronizes vehicle and drone operations with the objective of minimizing the total duration of the joint tour. Using a combination of valid inequalities, pre-processing, and other bound tightening strategies, the tractability of the proposed MIP formulation is enhanced to produce exact solutions to benchmark instances having up to 24 customers, the largest in the literature. Our work further enabled exact solutions for certain networks involving 32 customers, twice the size of instances solved in the extant literature at that point. Since commercial parcel delivery typically may involve 100 packages per route, Chapter 2 builds on the breakthrough in Chapter 1 and proposes an optimization-based variable neighborhood search heuristic. Starting with a relaxed problem structure, the heuristic uses two distinct MIP formulations to progressively integrate constraints relating to synchronous travel between the vehicle and drone as well as cyclic flights. By termination, the heuristic restores all problem assumptions to yield a high-quality feasible solution. On a set of benchmark instances from the literature, the heuristic improves upon the best-known results for 113/120 instances having up to 100 nodes, with comparable computational effort to approaches in the literature (Schermer et al., 2018). We also propose several pre-processing techniques that simplify decision-making associated with drone cycles, and analytically investigate the conditions under which they are optimal. Chapter 2 is based on the work of El-Adle et al. (2021). Building upon the previous work, Chapter 3 examines the idea of simplifying operational last-mile delivery problems that involve vehicles and portable drones. By optimizing a multi-period last-mile delivery problem, it is possible to identify, at a more tactical level, customers that should be served by drone and those who should receive their packages by vehicle in a given season. This tactical assignment is informed by geospatial and demand data analytics and an optimization model that optimizes the underlying operational vehicle routing and drone flights over the multi-period horizon under investigation. We propose a novel mixed-integer formulation to this challenging multi-period last-mile delivery problem with drone eligibility, which is also embedded in an optimization-based variable neighborhood search that effectively and consistently discovers near-optimal solutions (within 0.5% optimal) for networks involving 200 customers in manageable computational times.