The technology-enabled ride-pooling (RP) is designed as an on-demand feeder service to connect remote areas to transit terminals (or activity centers). We propose the so-called “hold-dispatch” operation strategy, which imposes a target number of shared rides (termed the ride-pooling size) for each vehicle to enhance RP’s transportation efficiency. Analytical models are formulated at the planning level to estimate the costs of the RP operator and the patrons. Accordingly, the design problem is constructed to minimize the total system cost concerning the system layout (i.e., in terms of service zone partitioning), resource deployment (i.e., fleet size), and operational decisions (i.e., RP size). The proposed models admit spatial heterogeneity arising from the non-uniformity of demand distributions and service locations, and can furnish heterogeneous designs. Closed-form formulas for the optimal zoning and fleet size are developed, which unveil fundamental insights regarding the impacts of key operating factors (e.g., demand density and distance to the terminal). Extensive numerical experiments demonstrate (i) the effectiveness of heterogeneous service designs and (ii) the advantage of the proposed RP service with hold-dispatch strategy over alternative designs studied in the literature, i.e., RP with a “quick-dispatch” strategy and flexible-route transit, in a wide range of operating scenarios. These findings can assist transportation network companies and transit agencies in successfully integrating RP and transit services. •Formulated analytic models for designing ride-pooling as an on-demand feeder to terminals.•The proposed model admits spatially non-uniform demand and accordingly furnishes heterogeneous designs.•Closed-form solutions are derived with new insights into the optimal designs.•The ride-pooling system outperforms the flexible-route transit with up to 6% cost saving.