Don't try to go into details without first acquiring a base in historical linguistics; and it obviously depends on your purpose:
If you are designing a constructed language intended to be used as an auxiliary intercommunication language then you generally consider it static.
If you are designing a language intended to stand for the naturally evolved language of a fictional linguistic group then you don't "mix [and] match [grammatical] features to find a balance you like", but rather you mimic the features of an attested language or two; and then you follow on with an immitation of attested processes of linguistic change.
For example, in the Indo-European family the general tendencies over the last three or four millennia were
Reduction and regularization of inflectional paradigms:
- Three numbers (singular, dual, plural, e.g. in Sanskrit and Ancient Greek) to only two numbers.
- Three grammatical genders (masculine, feminine, neuter) to only two grammatical genders, or even no grammatical gender (e.g., English or Persian).
- Many nominal case forms to few nominal case forms, or to one (as in English or French, where nouns are no longer declined for case).
- Many mood and tense verbal forms to few mood and tense forms, or, as in the case of English, very few.
- Elimination of "irregular" paradigms (which themselves were the remnants of old regular patterns) in favor of simpler patterns.
In parallel with the simplification of morphology, a shift from free word order (made possible by the rich morphology) to strict word order (to compensate for the morphological simplification). We say that ancient Indo-European language were more synthetic and modern languages more analytic.
This evolution in grammar was obviously superperposed with sound changes, and interconditioned by them; they cannot be separated.
The general trends are just that, trends. Occasionaly, in the history of a specific language, they could be reversed; for example, in French the future tense is a newly evolved synthetic form derived from an analytic form (infinitive + avoir) which had replaced the old Latin synthetic form, which itself was descended from an analytic form (verbal stem + *bhu)...
But all this is useful only if the work is ample enough to follow the evolution of one or more languages over millennia; for example, Tolkien did it with his Middle-Earth elves: Common Eldarin -- Quenya (and Telerin) -- Sindarin. But then Tolkien knew what he was doing: he was a professional historical linguist (he worked on the Oxford English Dictionary) and a professor of Anglo-Saxon literature.
TL;DR
In the Indo-European language family the general trends in grammatical evolution over the last 3 or 4 millenia were:
- Synthetic structure → analytic structure (with English exhibiting a trend towards a fully isolating language)
- Complex morphology → simple morphology → almost no morphology (e.g., English)
- Reduction in morphological categories (for example, English has almost completely lost the entire category of grammatical gender) (complicated by the constant emergence of new categories, which in turn become candidates for elimination; consider for example English verbal aspect)
- Elimination of irregular paradigms (complicated by the constant emergence of new paradigms, which in turn become irregular due to sound changes)
- Free word order → limited freedom → fixed word order
Omnis ars naturae imitatio est (all art is an imitation of nature). Seneca, Ad Lucilium Epistulae Morales, 65.3, published around 65 CE.