Depth perception is paramount to tackle real-world problems, ranging from\nautonomous driving to consumer applications. For the latter, depth estimation\nfrom a single image represents the most versatile solution, since a standard\ncamera is available on almost any handheld device. Nonetheless, two main issues\nlimit its practical deployment: i) the low reliability when deployed\nin-the-wild and ii) the demanding resource requirements to achieve real-time\nperformance, often not compatible with such devices. Therefore, in this paper,\nwe deeply investigate these issues showing how they are both addressable\nadopting appropriate network design and training strategies -- also outlining\nhow to map the resulting networks on handheld devices to achieve real-time\nperformance. Our thorough evaluation highlights the ability of such fast\nnetworks to generalize well to new environments, a crucial feature required to\ntackle the extremely varied contexts faced in real applications. Indeed, to\nfurther support this evidence, we report experimental results concerning\nreal-time depth-aware augmented reality and image blurring with smartphones\nin-the-wild.\n