There is no dichotomy between tone languages and intonation languages. The available evidence indicates that all languages have intonational systems. Some languages have lexical stress, some have lexical tone, and some appear to have no form of word "prominence" (Ethiopic Semitic, for example). Pitch-accent is a dubious category, which has largely been replaced with the concept of restricted-tone language (where only one syllable per word has a distinctive mark). You do find "gradual tone changes" at the sentence level in tone languages. The impression that you're getting probably arises from the practice of taking Mandarin Chinese to be typical of tone languages. Most of the time, the expression "tone languages, like ___" is usually completed with "Chinese" and not "Luganda".
The most relevant properties that realize systems of intonation and tone are F0, duration, amplitude, phonation, and formant values (listed in order from most-robust and common to least so). Formant values are generally not direct manifestations of a category contrast, but rather result from duration differences (target-undershoot in shorter vowels). The bottom-line form of formant differences is vowel change depending on the tone or stress property of a syllable, such as reduction of some unstressed vowels of English to schwa, or the vowel quality changes in the Min Chinese tone system.
Phonatory differences usually are parasitic off of F0 differences, with higher F0 correlating with creaky voice and lower F0 correlating with breathy voice, and these are usually garden-variety allophonic concomitants of stress (H tone) vs. unstressed (L tone). However, they can be phonologized and made independent, and especially in Southeast Asian languages tone like Vietnamese, they become part of the defining essence of a particular tone.
Amplitude is not an independently controllable property at the segment / syllable level, but differences can be observed which correlate with duration and F0. Amplitude is potentially exploitable as an aspect of intonation, where you can "breathe harder" thus talk louder when exclaming "I told you to bring the metallurgical kit", though there will be F0 differences as well. The F0 / amplitude correlation exists in tone and non-tone languages alike.
These three properties are not particularly robust as correlates of lexical stress versus tone across languages. One gains more traction with duration and F0. F0 differences are probably the most common specific correlate of the stressed / unstressed difference and of course are the quintessence of tonal contrast, and duration differences also often are associated with the stressed / unstressed contrast – but not generally with tone. Looking at phonological processes, this is common-enough that we have names for it: the Weight-to-Stress principle where stress is assigned to heavy syllables, and the Stress-to-Weight principle, where stressed syllables get lengthened. There is a weak tonal correlate of the former, where tone is sometimes attracted to heavy syllables, there does not seem to be an active Tone-to-Weight principle where syllables bearing a H tone, for example, get lengthened. (Confusion often arises over the fact that tone languages often impose a phonological limit on the number of tones allowed on short syllables, so that contours are only possible on bimoraic syllables, but that is a separate phenomenon). From the phonological POV, it is interesting that stress can induce lengthening, but tone apparently cannot.
It is pretty much impossible to identify a language as tonal versus non-tonal just based on acoustic properties. Theoretically, you might be able to identify outliers based on a large-corpus acoustic study of certain languages. The variability of F0 over time in Cantonese is rather high, and you could use that to determine that Cantonese is probably a tone language, whereas the F0 variability of Tigre is low and you could thus suspect that it is not a tone language. The search for the Holy Grail of finding necessary or sufficient acoustic diagnostics of tone vs. stress has been called off.